Skip to main content
    Back

    Delivery

    From evaluation to production

    This page describes how a deployment is carried out, which contract models are available, where the boundary lies between the locked engine and your own configuration, and how the copilot is embedded into your surfaces.

    Delivery

    Four steps to go-live

    The engine is unchanged between deployments. The work in a new deployment sits in the configuration layer on top, which limits both lead time and the scope of a renewed review.

    01

    Knowledge

    Search is pointed at your documentation, your policies and your product pages. Nothing outside that scope is answerable.

    02

    Persona

    Name, face, opening line and industry glossary are set per customer and versioned.

    03

    Actions

    A short, defined list of what the copilot may offer. Nothing runs until the customer presses it themselves.

    04

    Embed

    A script tag on your own domain, or the SDK inside your app. Spend caps and escalation recipients are set before launch.

    Business model

    Three contract forms depending on how much of the surface you own

    Licensed copilot

    We run the engine. You get your own copilot on your own domain, with your knowledge base and your brand.

    White label

    Your name up front, our engine behind it. Suits platforms and groups that resell to their own customers.

    Embedded engine

    The engine inside your product, with your own operations team handling configuration and escalation.

    Boundary of responsibility

    What is fixed, and what is yours

    The line is drawn deliberately. Everything that affects security, cost and consequence is fixed in the engine and identical in every deployment. Everything that makes the copilot yours is configuration, not new development.

    Fixed in the engine

    The same code in every deployment

    This is why it can be reviewed once and trusted across markets. No tenant-specific exceptions in the decision logic.

    • Language handling and fallback behaviours
    • Conversation styles and the tone contract
    • Topic boundary, abuse protection and cost caps
    • Escalation rules and audit trail
    • The handover format and duplicate protection
    • Response delivery, retries and wait states
    • Feedback capture and quality logging

    Configured per customer

    Your knowledge, your tone

    A new market or a new business area is a package, not a new project. This keeps lead time and running cost down.

    • Knowledge sources and search scope
    • Persona: name, face, opening line
    • Industry glossary for the business
    • The list of actions the copilot may propose
    • Escalation recipients and opening hours
    • Which channels and surfaces are activated
    • Brand, placement and embed surface

    Integration

    One script tag. No core system.

    The copilot loads as an isolated client on your domain. Knowledge base, language pack and CTA map are governed by configuration, not by a new platform release.

    No backend integration
    The copilot makes no calls to core systems.
    No customer data
    No transactions, no personal data in the engine.
    Per-tenant spend caps
    Budget and module access are set in the configuration.
    index.html
    <script
      src="https://cdn.novacopilot.ai/v1/copilot.js"
      data-tenant="your-tenant-id"
      data-market="sv-SE"
      data-modules="cards,payments,claims"
      defer
    ></script>
    Streaming responseAudit log per moduleEU region
    Nova, digital assistantNovaCopilotLegal assistant